Mining accident detection using machine learning methods

Mining activity carries inherent risks in its work These risks have produced many accidents in Chilean and all mining history, some of them with fatal consequences Does the state and the environment of the mine affect workers performance and security? Do long periods without accidents generate overconfidence in workers? Do recent accidents generate insecurity in workers? These type of questions sought to be answered by the Chilean consultancy SolMat using mathematical modeling to generate a computational tool that will allow to anticipate the occurrence of accidents in order to improve the safety of workers in Minera Los Pelambres This paper generates and validates predictive models for daily and weekly prediction of accidents in all productive sectors as one large sector, and on a segmentation of the whole productive place into three specific sectors Solmat used Machine Learning techniques with supervised training, obtaining with independent testing bases results of 70% of total accuracy for the job, and 75%, 85% and 75% of accuracy for the 3 previous segmentations, being able to detect more than the half of accidents in each case For the daily case, accuracy is similar, but with less accident detection